Tool Retrieval on TOOLRET Zero-Shot Macro-Avg
36.9nDCG@10TOOLQP
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| TOOLQPMethod Category=Fine-tuning, Training Data Size=10k, Backbone Model=1.7B2026.01 | 36.9 | 35.8 | |
| bge-gemmaMethod Category=Re-ranking with cross-encoder2026.01 | 34 | 29.7 | |
| Q2P/SFTMethod Category=Fine-tuning, Training Data Size=10k, Backbone Model=1.5B/1.7B models2026.01 | 33.2 | 31.1 | |
| Q2E/ZSMethod Category=Prompting, Backbone Model=Qwen3-30B-A3B-Instruct2026.01 | 32 | 29.8 | |
| gte-Qwen/ContrastiveFTMethod Category=Fine-tuning, Training Data Size=10k, Backbone Model=1.5B/1.7B models2026.01 | 31.5 | 29 | |
| Q2D/PRFMethod Category=Prompting, Backbone Model=Qwen3-30B-A3B-Instruct2026.01 | 31.2 | 27.9 | |
| Q2E/PRFMethod Category=Prompting, Backbone Model=Qwen3-30B-A3B-Instruct2026.01 | 31 | 28.4 | |
| Re-InvokeMethod Category=Prompting, Backbone Model=Qwen3-30B-A3B-Instruct2026.01 | 30.3 | 27.9 | |
| bge-m3Method Category=Re-ranking with cross-encoder2026.01 | 30.1 | 26.8 | |
| Base Retriever (gte-Qwen)Method Category=Base Retriever2026.01 | 29.8 | 27 | |
| D2QMethod Category=Prompting, Backbone Model=Qwen3-30B-A3B-Instruct2026.01 | 28.4 | 26.4 | |
| HyDE/PRFMethod Category=Prompting, Backbone Model=Qwen3-30B-A3B-Instruct2026.01 | 27.6 | 24.4 | |
| Q2D/ZSMethod Category=Prompting, Backbone Model=Qwen3-30B-A3B-Instruct2026.01 | 26.7 | 26.9 | |
| TOOLQP-FORMATMethod Category=Fine-tuning, Training Data Size=10k, Backbone Model=1.7B2026.01 | 26.1 | 25.9 | |
| HyDE/ZSMethod Category=Prompting, Backbone Model=Qwen3-30B-A3B-Instruct2026.01 | 19.9 | 21.8 |